Discussion

This study revealed several important trends regarding EEG utilization and diagnostic yield. The proportion of EEG showing abnormalities (∼40%) falls within the ranges previously reported in the literature. ^{Reference Angus-Leppan8–Reference Gilbert and Gartside14} As observed in a previous study, ^{Reference Phamnguyen, Szekely and Swinburn11} abnormal findings were more likely in patients undergoing EEGs in inpatient and emergency settings compared to outpatients, likely reflecting greater illness severity and a higher pretest probability of neurological pathology.

Clinical indication was strongly associated with diagnostic yield. Indications such as neuro-prognostication and suspected non-convulsive status epilepticus were highly predictive of EEG abnormalities, which is expected given that patients referred for these indications were acutely ill or in critical care. Conversely, the indication of “syncope or presyncope” was associated with a significantly reduced likelihood of abnormalities (Table 3), supported by literature ^{Reference Mecarelli, Pulitano, Vicenzini, Vanacore, Accornero and De Marinis22–Reference Poliquin-Lasnier and Moore24} and NICE/SIGN guidelines, ^25,26 showing low yield in EEGs referred for syncope. These findings reinforce the need for careful patient selection and indication-based requisition to optimize resource use.

The NICE guidelines recommend that EEGs be primarily requested to support the diagnosis of epilepsy and differentiate between seizure types. ²⁵ Additionally, EEG is not recommended to investigate transient loss of consciousness. ²⁷ NICE guidelines should be strictly adhered to maximize EEG potential and decrease inappropriate referrals. ²⁵ Similarly, the SIGN guidelines state that an EEG is recommended when there is a high clinical suspicion of epilepsy and should not be performed if a patient’s symptoms point to non-epileptic events, such as syncope. ²⁶ Furthermore, EEG should not be the only tool used to diagnose epilepsy. As for Choosing Wisely Canada, there are five neurology recommendations, one of which states, “Don’t order neuroimaging or EEG in asymptomatic patients in the emergency department with syncope and a normal neurological evaluation.” ²⁸

Guidelines for EEG utilization vary internationally, reflecting differences in scope and level of specificity. For example, NICE and SIGN guidelines recommend EEG primarily to support the diagnosis of epilepsy and to help differentiate between seizure types. ^25,26 In contrast, Choosing Wisely Canada provides broader recommendations focused on reducing unnecessary testing in situations such as syncope. NICE and SIGN guidelines offer more detailed, indication-specific guidance, while Choosing Wisely Canada emphasizes resource stewardship.

These variations influence diagnostic approaches, test indications and overall management strategies involving EEGs. Countries urgently need to develop and establish clinical guidelines for EEG utilization to decrease the number of referral requests, as previous research has suggested that 50–60% of requests are unnecessary. ^{Reference Smith, Bartolo, Pickles and Tedman29} Given the increasing demand and labor shortage, guidelines can help promote the optimal use of EEG services to decrease the burden on the healthcare system.

Multivariate regression analyses provided further insight into predictors of EEG abnormalities. Older age, inpatient status and use of anti-seizure or sedative medications significantly increased the odds of detecting any abnormality. In contrast, predictors of epileptiform abnormalities included some distinct predictors like younger age and history of epilepsy. This age-dependent pattern aligns with known developmental differences in seizure susceptibility and EEG findings. ^{Reference Monif and Seneviratne10,Reference Gilbert and Gartside14} Several studies have identified younger age as a factor associated with higher diagnostic yield in EEG, even when analyzing predominantly adult populations. ^{Reference Monif and Seneviratne10,Reference Marsan and Zivin18} However, the significance of age as a predictor differs from other studies conducted with varying cohorts that found no effect, including one with exclusively an inpatient population and one with patients having a known history of seizures. ^{Reference Phamnguyen, Szekely and Swinburn11,Reference Betjemann, Nguyen, Santos-Sanchez, Douglas and Josephson17,Reference Sundaram, Hogan, Hiscock and Pillay19} This presents the possibility that the correlation of age with the likelihood of abnormalities may be impacted by other confounding factors, such that inpatients are usually older.

Inpatient status is another variable associated with higher diagnostic yield. ^{Reference Phamnguyen, Szekely and Swinburn11} Inpatient EEGs are frequently performed for acute conditions, such as possible seizures or cognitive/behavioral changes, where abnormalities are more likely. Sedating medications, commonly used in inpatient intensive care settings, may also influence diagnostic yield but have not yet been explored in this context. Research suggests that sedatives – such as propofol – alter EEG patterns by inducing changes in frequency and waveform characteristics, causing burst suppression and generalized slowing of EEG. ^{Reference San-juan, Chiappa and Cole30,Reference Sonkajarvi, Puumala and Erola31} These effects can mask or mimic pathological findings, potentially complicating EEG interpretation.

A prior history of epilepsy has also shown to be a predictor of higher diagnostic yield. ^{Reference Monif and Seneviratne10,Reference Phamnguyen, Szekely and Swinburn11} Patients with known epilepsy are more likely to exhibit epileptiform activity on EEG, as their condition involves recurrent abnormal brain activity, which can persist despite treatment. However, ASMs can complicate this relationship. While ASMs are designed to suppress seizures and epileptiform activity, their effects on EEG yield are variable. Some studies suggest that ASMs significantly reduce the likelihood of detecting abnormalities, whereas others have found no correlation. ^{Reference Monif and Seneviratne10,Reference Marsan and Zivin18,Reference Sundaram, Hogan, Hiscock and Pillay19} One study in a pediatric cohort even showed that odds of detecting an abnormality were greater when the patient was on ASMs, which may reflect that ASM use correlates with a more severe or established pathology. ^{Reference Gilbert and Gartside14}

Importantly, yield should not be interpreted as a proxy for clinical utility. In this study, “diagnostic yield” reflects the presence of abnormalities detected by the test. It does not directly indicate whether these findings meaningfully influence diagnosis, treatment decisions or patient outcomes. For instance, an abnormal EEG in a patient with known epilepsy already established on treatment would contribute to diagnostic yield irrespective of whether it leads to a change in treatment. In a study evaluating the utility of the routine EEG in the emergency department, an EEG was considered “useful” only when it led to changes in clinical management. ^{Reference Rodriguez Quintana, Bueno, Zuleta-Motta, Ramos, Velez-van-Meerbeke and tNR32} Future prospective studies are therefore needed to assess the extent to which EEG findings influence management across clinical settings.

We compared distinct patient groups who had either routine or sleep-deprived EEGs and found no significant difference in epileptiform yield. The impact of sleep deprivation as an activation procedure has been inconsistently reported. Some studies demonstrate a clear benefit, particularly in patients with suspected epilepsy. ^{Reference DeRoos, Chillag, Keeler and Gilbert15,Reference Giorgi, Perini and Maestri33} However, others find negligible differences compared to routine EEGs, which may reflect variations in protocols or patient populations. ^{Reference Daniel, Dennis, Anita and Roy34,Reference Guerrero-Aranda, Enriquez-Zaragoza, Lopez-Jimenez and Gonzalez-Garrido35}

As sleep-deprived EEGs are often performed following a routine EEG in the same patient, we also conducted a within-subject analysis. This follow-up similarly showed no significant increase in epileptiform abnormalities, although the small sample size may limit the strength of this finding. Still, the clinical practice of using routine EEGs to guide decisions for subsequent sleep-deprived studies appears justified in select cases, especially given that a third of these follow-up tests were neurologist-recommended based on prior results. Interestingly, the lack of association between achieving N2 sleep and the detection of epileptiform abnormalities is consistent with literature suggesting that sleep-deprived EEGs may increase interictal epileptiform discharges even during wakefulness, independent of capturing sleep stages. ^{Reference Degen36,Reference Fountain, Kim and Lee37}

Over half (56%) of the EEG requisition forms in this study were found to be incomplete. The “Medication” and “Patient History” sections show the highest completion rates, 95.6% and 98.8%, respectively. In contrast, the remaining three sections have comparatively low completion rates, with many requisitions containing only one out of these three sections completed. This pattern highlights inconsistencies in how the form is filled out and suggests the current requisition form may lack a clear structure. If information is missing from the requisition form, healthcare professionals reviewing the forms and reports are often required to seek out additional clinical details, which is time-consuming and inefficient. ^{Reference Laing, Jarmain and Elliott38} Consequently, the data points to a broader issue regarding the overall poor quality of the form. Clinicians completing these requests may be uncertain about what information is required in each section due to ambiguous and overlapping section titles of the current requisition form. This analysis also identified important differences in requisition completeness across patient settings and ordering specialties. Compared with outpatient requests, requisitions originating from the emergency department and critical care were significantly less likely to be complete, consistent with research indicating that under time pressure, physicians’ documentation accuracy declines. ^{Reference Mashoufi, Ayatollahi, Khorasani-Zavareh and Talebi Azad Boni39,Reference Soucek, Reggiani and Kairies-Schwarz40}

When examining ordering physicians’ specialties, neurologists were chosen as the reference group in the logistic regression because research has shown neurologists tend to submit a high proportion of “appropriate referrals.” ^{Reference Pearce and Cock41} However, in the present study, neurologists were less likely to submit complete requisitions compared with physicians in all other specialties. This pattern may reflect other forms of inter-departmental communication or variability in how neurologists balance clinical knowledge with form completion. As well, verbal discussion between physicians about the patient may partially substitute written communication on the paper requisition. Physicians with less knowledge of neurology may provide additional less important details or be less aware of what information an EEG can provide.

A clear, effective and well-documented requisition form improves the accuracy and efficiency of test interpretation and ensures that diagnostic tests are appropriately tailored. ^{Reference Hartveit, Vanhaecht, Thorsen, Biringer, Haug and Aslaksen21} However, previous literature has found that referral guidelines lack consistency and specificity, therefore reducing the number of quality referrals in Canada. ^{Reference Laing, Jarmain and Elliott38} A review of Canadian referral guidelines identified 42 distinct types of referral data recommended for inclusion in referral documentation. However, there was substantial variability in how consistently these data elements were included across referral guidelines, with individual items appearing in only 8 to 77% of recommendations. Although these findings were derived from a variety of referral processes and not specific to a requisition form for a single diagnostic test, the results highlight broader systemic inconsistencies. This variation highlights a notable discrepancy between standardized guideline recommendations and referral forms, suggesting inconsistency in referral processes and potential barriers to efficient communication. ^{Reference Laing, Jarmain and Elliott38} Therefore, high-quality requisition forms remain important for enhancing communication within healthcare systems.

Incomplete requisitions are a persistent issue in broader healthcare settings and are associated with a range of adverse consequences. Studies in other diagnostic settings, including a study of incomplete laboratory request forms, have shown that incomplete request forms can contribute to errors in patient scheduling, miscommunication and increased risk of diagnostic errors. ^{Reference Nutt, Zemlin and Erasmus42} Inadequate information on referral forms can also lead to repeated testing, unnecessary procedures or missed diagnoses compromising patient outcomes and highlighting the importance of accuracy and thorough documentation. ^{Reference Savoy, Khazvand and Mathew43–Reference Kotze and Parrish45} In this context, the inclusion of a clearly defined clinical question is particularly important, as it helps ensure that diagnostic testing is appropriately targeted and that results are interpreted within the correct clinical framework. While some research suggests that initial EEG interpretation may be performed without clinical information to reduce history bias, this step is typically followed by integration with clinical context to ensure meaningful interpretation. ^{Reference Amin and Benbadis46,Reference Nascimento, Jing and Beniczky47} Standardized reporting guidelines from the American Clinical Neurophysiology Society emphasize that clinical correlation is an integral component of EEG interpretation. ^{Reference Tatum, Olga and Ochoa48} Therefore, EEG requisition forms should explicitly include a specific clinical question to improve the contextual framework in which the test is interpreted and reported. The results of this study supported the updating of KHSC’s EEG requisition form (Supplementary Figure 2) to address unnecessary duplication in the sections and provide clear indications for the investigation.

On February 25, 2025, the provincial health authority Shared Health notified all Manitoba physicians that an incomplete MRI requisition would be rejected ⁴⁹ as they had noticed a trend in incomplete physician requisitions. Although this policy pertains to MRI rather than EEG, it serves as an example of a system-level intervention aimed at improving requisition completeness. A similar approach could be considered for EEG requisitions as a potential future strategy to improve the quality of clinical information provided.

The low diagnostic yield seen in outpatient settings and from family physicians and pediatricians may suggest potential overuse or inappropriate use of EEG. Targeted educational interventions may help address this issue by enhancing clinicians’ understanding of appropriate referral. ^{Reference Smith, Bartolo, Pickles and Tedman29} Other factors, including variability in EEG acquisition practices, laboratory resources and clinician expertise, may further moderate the extent to which requisition quality translates into improved diagnostic outcomes. As such, optimizing requisition forms should be considered as one component within a wider framework of improving EEG service delivery and resource utilization. By focusing on these areas of weakness, it is possible to increase diagnostic efficiency, reduce unnecessary testing and ultimately improve patient outcomes.

This study should be interpreted in the context of several limitations. First, the data were collected over a limited sampling period from a single tertiary-care center, which may restrict the generalizability of the findings to other institutions with different referral practices, patient populations or requisition formats. Second, the high proportion of incomplete requisitions limits the generalizability of our findings regarding the quality of clinical information conveyed through requisition forms, as this may reflect site-specific documentation practices or form design. Additionally, several variables may not have been entirely independent, introducing collinearity. However, these were accounted for through multivariate regression analysis to minimize confounding effects.

Another limitation relates to the source of clinical information. Clinical variables, including ASM use and epilepsy diagnosis, were obtained solely from EEG requisition forms and reports rather than comprehensive medical charts, which may have introduced under-ascertainment bias. Furthermore, we did not have access to full clinical charts or longitudinal follow-up data to determine whether EEG results influenced diagnosis or patient management, as our referrals come from all specialists and family doctors in the region. As a result, the impact of EEG findings on clinical decision-making could not be assessed.

Yield may also have been affected by the time between the event and the test being performed. Evidence suggests that EEG timing after a first unprovoked seizure affects its diagnostic and prognostic value. In one study evaluating EEG timing following a first unprovoked seizure, both early and late EEGs were found to be valuable but served different purposes: early EEGs (within 2 days) were more useful for diagnostic evaluation, whereas later EEGs provided greater prognostic value in predicting seizure recurrence. ^{Reference Gilboa, Hanuka, Shifman and Ekstein50} In our institution, most inpatient requests are performed on the same day, whereas the wait time for routine outpatients was under 2 months at the end of the study period. Urgent outpatient requests (e.g., first seizure, possible infantile spasms) were prioritized. However, the interval between the clinical event and EEG acquisition was not consistently recorded in requisitions, and therefore its impact on diagnostic yield could not be systematically analyzed in this study.

Another limitation to this study is that, in Cohort 1, only the first EEG within the study period was retained for analysis to maintain independence of observations. However, this approach may not capture longitudinal diagnostic information. Patients may have undergone additional EEGs prior to, within or following the study period, and it is possible that abnormalities detected on these recordings were not reflected in the analysis. As a result, the reported diagnostic yield reflects the first recorded EEG within the study time frame rather than the first lifetime EEG or the full spectrum of EEG results across repeated testing within individuals.

During the study period, all EEGs required paper-based free-text requisitions. However, the widespread adoption of electronic medical records could limit the quality/quantity of information provided in the future if free-text boxes are replaced by a drop-down box with limited options. Lastly, out of the 1215 total completed EEGs, only 960 requisition forms were available, limiting the cohort size. Future research may focus on optimizing EEG indications to improve diagnostic efficiency and yield, as well as refining clinical guidelines to best suit specific populations. Prospective studies incorporating longitudinal patient follow-up may also be valuable to better evaluate how EEG results influence diagnostic outcomes and patient management.